Mikey, There's a nasty rumor running around that you may have solved an issue I'm having and am fixin' to make worse... I switched a box that had been running a RX 580 and then a RX 6600 over to running a RTX 3060ti with much help from mountkidd.
This is all on a Linux Mint v20.3 with the 5.15 kernel installed.
Linux mint doesn't use an xorg.conf (I'll not pretend to understand the details behind this).
Upon removing all traces of AMDGPU and installing the v530 Nvidia driver it built a xorg.conf which promptly led to a black screen. Mountkidd has been as patient as possible but I don't have a working xorg.conf.
Likely related is that this box has a Ryzen 7 5700G and the monitor is plugged up the iGPU which is exclude from BOINC via the appropriate cc_config.xml '<ignore ati dev 1' (an approximation of that line). This all works fine and the RTX 3060ti is happily doing WUs but it seems control options are lacking that would require a working xorg.conf.
I was able to install the needed driver from ROCm with amdgpu-install_5.4.50403-1_all.deb
from AMD's site but looking at the apt sources it seems to target only one version number. Is there a way to track stable updates as they appear? Thank you.
I saw a post that said with the RX 6600 GPU running more than one at a time didn't speed up Gamma Ray Burst work units. Is the same true of Gravitation Wave units? Thank you.
If you read it again you will find the comment was about MeerKAT tasks and not the Gamma Ray Binary Pulsar search (FGRPB1G). Running FGRPB1G tasks x2 should give a reasonable improvement in output (5 - 10% perhaps), particularly if you take care not to have the pair starting and finishing simultaneously.
However, this could easily be masked (or significantly reversed) by CPU searches if your CPU is being overloaded. If you see longer than normal run times (or big variations from task to task) it's a sign of an overloaded CPU.
For GW tasks, the latest CPU versions seem to have extremely large memory requirements. I don't know if that translates to the GPU versions as well. If you want to run GW GPU tasks, just run a few at x1 (say 10) and see what the average run time turns out to be. Then run another batch at x2 and as long as the average per task run time is less than twice what it was for x1, then you are winning. Your GPU shows as having 8GB so you should be able to run at least x2 hopefully.
To be aware of things like changes in memory requirements, etc., it's useful to follow the progress for new (or changed) searches in the relevant thread in Tech News.
I saw a post that said with the RX 6600 GPU running more than one at a time didn't speed up Gamma Ray Burst work units. Is the same true of Gravitation Wave units? Thank you.
My RX 6600s (8GB Sapphires) return a better through put using 2x on the 'Mad Dog' GW WUs. I have all 3 types of WUs (BRP, FGRP and GW) set to run 2x via the Prefs-> Project page and disposed of the app_config file. This is under Linux Mint v20.3.
I don't recall the "GPU utilization factor" I had on non-GW WUs when these were all RX 580 cards but I know the GW WUs were also 2x then and also showed better through-put but were pushing max memory on my 4GB Sapphire cards.
AMD does publish about every month a new version of GPU-driver. If I look into the published advances, I see only such for games, I'm not interested in. I'm riding a RX6600 successfully without any problems from start on.
So, my question: Does it give remarkable advantage for our work to install every new driver version?
AMD does publish about every month a new version of GPU-driver. If I look into the published advances, I see only such for games, I'm not interested in. I'm riding a RX6600 successfully without any problems from start on.
So, my question: Does it give remarkable advantage for our work to install every new driver version?
Kind regards and happy crunching
Martin
I have only run NVIDIA GPUs, never owned or use an AMD GPU. With that said, I would say no to installing the latest driver. The reason as far as NVIDIA goes, the latest drivers are geared for gamers. And since you say your not interested in games (and neither am I) they wouldn't be beneficial to you.
My personal opinion on AMD GPU drivers - I would think that they're also geared towards gamers. If you can read any info regarding the latest drivers and find out what they're good for, that would be beneficial to you. Only get the ones that are correcting something or modifying something that YOU personally are experiencing.
I am running an rx 6600 8GB but haven't done 2x on the GW work units. Radeontop shows 4.8GB with some work units so it shouldn't work with only 8GB running two. Are you sure you are not getting errors?
I am running an rx 6600 8GB but haven't done 2x on the GW work units. Radeontop shows 4.8GB with some work units so it shouldn't work with only 8GB running two. Are you sure you are not getting errors?
Frankly I don't know how this works... but I'll be glad to make something up... it's almost as if the driver/app when doing a malloc/getmain memory request is adjusting amount of memory requested to what is on the card (I'm sure this has limits) but I ran 2x MDGW WUs on my 4GB RX 580 cards and later on my 8GB RX 6600 cards w/o memory issues.
Skip
PS: Corectrl will show what each WU has allocated once running and total for each card and the overhead for xorg, cinnamon, etc if the card was also driving a monitor. I've never used Radeontop.
You can see the host
)
You can see the host https://einsteinathome.org/host/13135091
It has 2 RX 6600s in it but as of this morning it's only running one of them. Running
2xon allon mostWUs.Skip
Just the guy I was looking
)
Just the guy I was looking for ;-)
Mikey, There's a nasty rumor running around that you may have solved an issue I'm having and am fixin' to make worse... I switched a box that had been running a RX 580 and then a RX 6600 over to running a RTX 3060ti with much help from mountkidd.
This is all on a Linux Mint v20.3 with the 5.15 kernel installed.
Linux mint doesn't use an xorg.conf (I'll not pretend to understand the details behind this).
Upon removing all traces of AMDGPU and installing the v530 Nvidia driver it built a xorg.conf which promptly led to a black screen. Mountkidd has been as patient as possible but I don't have a working xorg.conf.
Likely related is that this box has a Ryzen 7 5700G and the monitor is plugged up the iGPU which is exclude from BOINC via the appropriate cc_config.xml '<ignore ati dev 1' (an approximation of that line). This all works fine and the RTX 3060ti is happily doing WUs but it seems control options are lacking that would require a working xorg.conf.
What 'cha know 'bout this that you could share?
Thanx, Skip
I was able to install the
)
I was able to install the needed driver from ROCm with amdgpu-install_5.4.50403-1_all.deb
from AMD's site but looking at the apt sources it seems to target only one version number. Is there a way to track stable updates as they appear? Thank you.
I saw a post that said with
)
I saw a post that said with the RX 6600 GPU running more than one at a time didn't speed up Gamma Ray Burst work units. Is the same true of Gravitation Wave units? Thank you.
Ben Scott wrote:... running
)
I guess you may be referring to this comment.
If you read it again you will find the comment was about MeerKAT tasks and not the Gamma Ray Binary Pulsar search (FGRPB1G). Running FGRPB1G tasks x2 should give a reasonable improvement in output (5 - 10% perhaps), particularly if you take care not to have the pair starting and finishing simultaneously.
However, this could easily be masked (or significantly reversed) by CPU searches if your CPU is being overloaded. If you see longer than normal run times (or big variations from task to task) it's a sign of an overloaded CPU.
For GW tasks, the latest CPU versions seem to have extremely large memory requirements. I don't know if that translates to the GPU versions as well. If you want to run GW GPU tasks, just run a few at x1 (say 10) and see what the average run time turns out to be. Then run another batch at x2 and as long as the average per task run time is less than twice what it was for x1, then you are winning. Your GPU shows as having 8GB so you should be able to run at least x2 hopefully.
To be aware of things like changes in memory requirements, etc., it's useful to follow the progress for new (or changed) searches in the relevant thread in Tech News.
Cheers,
Gary.
Ben Scott wrote:I saw a
)
My RX 6600s (8GB Sapphires) return a better through put using 2x on the 'Mad Dog' GW WUs. I have all 3 types of WUs (BRP, FGRP and GW) set to run 2x via the Prefs-> Project page and disposed of the app_config file. This is under Linux Mint v20.3.
I don't recall the "GPU utilization factor" I had on non-GW WUs when these were all RX 580 cards but I know the GW WUs were also 2x then and also showed better through-put but were pushing max memory on my 4GB Sapphire cards.
Hope this helps,
Skip
Hallo!AMD does publish
)
Hallo!
AMD does publish about every month a new version of GPU-driver. If I look into the published advances, I see only such for games, I'm not interested in. I'm riding a RX6600 successfully without any problems from start on.
So, my question: Does it give remarkable advantage for our work to install every new driver version?
Kind regards and happy crunching
Martin
astro-marwil
)
I have only run NVIDIA GPUs, never owned or use an AMD GPU. With that said, I would say no to installing the latest driver. The reason as far as NVIDIA goes, the latest drivers are geared for gamers. And since you say your not interested in games (and neither am I) they wouldn't be beneficial to you.
My personal opinion on AMD GPU drivers - I would think that they're also geared towards gamers. If you can read any info regarding the latest drivers and find out what they're good for, that would be beneficial to you. Only get the ones that are correcting something or modifying something that YOU personally are experiencing.
Proud member of the Old Farts Association
I am running an rx 6600 8GB
)
I am running an rx 6600 8GB but haven't done 2x on the GW work units. Radeontop shows 4.8GB with some work units so it shouldn't work with only 8GB running two. Are you sure you are not getting errors?
Ben Scott wrote:I am
)
Frankly I don't know how this works... but I'll be glad to make something up... it's almost as if the driver/app when doing a malloc/getmain memory request is adjusting amount of memory requested to what is on the card (I'm sure this has limits) but I ran 2x MDGW WUs on my 4GB RX 580 cards and later on my 8GB RX 6600 cards w/o memory issues.
Skip
PS: Corectrl will show what each WU has allocated once running and total for each card and the overhead for xorg, cinnamon, etc if the card was also driving a monitor. I've never used Radeontop.
PPS: Yea, I'm sure.
PPPS: Here's running 2x MDGWs on a slow 4GB RX 560: MDGW 2x or 4GB RX 560